Is a free GEO skill from GitHub enough, or do I need an MCP server?
For a one-off GEO audit of a site you know, an open-source Claude Code skill such as geo-seo-claude is often enough: it fetches pages, reads robots.txt rules for AI crawlers, drafts llms.txt and schema and writes a report. An MCP server earns its place when you need the same checks every run, live requests as each bot, measurement of what assistants say, and the same tools in Cursor or Codex. The question is who maintains the checks.
The short answer
A skill is instructions plus scripts that Claude Code follows. An MCP server is a program with tools that return the same structure on every call. A skill can do almost anything a server does if someone writes the scripts, so the real differences are where it runs, who keeps the checks current, and whether a run today can be compared with a run next month.
If you want one audit of your own site and you’re happy to read Python, start with a skill from GitHub. If you want checks after every deploy, live tests as each AI bot, and a record of what assistants say, you want a server, or your own scripts that do the same.
What the popular skills do
Read on each repository’s page on 15 October 2026.
| Skill | What it does | Install and needs | License | Limits we noticed |
|---|---|---|---|---|
| geo-seo-claude (Zubair Trabzada), about 11k stars | /geo with modes for audit, crawlers, llmstxt, schema, citability, brand mentions, content, report and PDF report; 5 parallel subagents; Python scripts fetch the home page and crawl the sitemap |
Python 3.10+, Claude Code, Git | MIT | Checks robots.txt for 14+ AI crawlers; the README doesn’t say it sends live requests as each bot or asks assistants |
| aaron-marketing-skills (Aaron He Zhu), SEO and GEO part | 16 skills in four stages: keyword research, competitor and SERP analysis, a content writer, schema builder, technical and on-page checkers, a rank tracker | /plugin marketplace add aaron-he-zhu/aaron-marketing-skills in Claude Code |
Apache-2.0 | The standalone seo-geo repo is frozen at v9.9.12; which data sources each skill uses isn’t spelled out on the page |
Sources: geo-seo-claude, seo-geo-claude-skills.
Skill or MCP server: what actually differs
| Skill (SKILL.md) | MCP server | |
|---|---|---|
| Runs in | Claude Code (some installers copy skills to other agents) | Any MCP client: Claude Code, Cursor, VS Code, Windsurf, Codex CLI, Gemini CLI |
| How steps run | The model reads the instructions; steps can vary between runs | The tool’s code runs the steps; the model decides when to call it |
| Results over time | Whatever files the scripts write | Usually a fixed format, so the next run can be compared |
| Outside data | What its scripts fetch with your keys | The same, plus anything the vendor runs on its side |
| Cost | Usually nothing | Nothing to a subscription |
| Trust | Runs with your permissions; read the scripts | Same for local servers; remote ones send data to the vendor |
Claude Code’s docs say skills live in .claude/skills/<name>/SKILL.md and can run bundled scripts (Claude Code docs, “Skills”, read 15 Oct 2026). Nothing stops a skill from calling an MCP server’s tools, so the two combine well.
Example: test any skill or server the same way
The check where tools differ most is crawler access. A robots.txt rule is a request; your CDN decides first. On 29 September 2026, katman.pro’s robots.txt allowed GPTBot while a Cloudflare setting returned 403 to it (own measurement). Compare the declared rule with what a request gets:
SITE=https://yoursite.com
# the declared rule
curl -s $SITE/robots.txt | grep -iA2 'OAI-SearchBot'
# what a request with each bot's user agent gets
for b in OAI-SearchBot/1.0 ChatGPT-User/1.0 Claude-SearchBot/1.0 PerplexityBot/1.0; do
echo "$b: $(curl -s -o /dev/null -w '%{http_code}' -A "Mozilla/5.0 (compatible; $b)" $SITE/)"
done
A spoofed user agent isn’t the real bot, and CDNs that verify bots by IP may treat it differently. So a 200 is a good sign, and a 403 is worth fixing: see Cloudflare blocking AI crawlers. Whatever you install, ask it to run this and check that its verdict matches. The AI crawler checker does the same from our side at no cost.
When to pick which
- One audit of your own site, or audits you sell to clients: a skill. geo-seo-claude even has prospect and proposal modes.
- You also work in Cursor, VS Code or Codex: an MCP server, since one config serves every tool.
- Checks after every deploy and a comparable record over months: a server, or a skill whose scripts write a fixed format you trust.
- Measuring what ChatGPT, Claude, Gemini and Perplexity say: something that calls them on a fixed question set and keeps the answers. Neither skill above describes that.
- Fixes in the code: either, as long as each job has an acceptance check you run yourself.
Where Katman fits
Katman MCP is a paid server, not a skill. Its 17 tools are listed on Tools: a code scan, a live-site audit through your CDN, fix prompts P0 to P14 with acceptance checks, page briefs, generated robots.txt, llms.txt and JSON-LD, and visibility runs that ask five assistants on your own OpenRouter key (about $2.40 a run on the default panel). It keeps results in .katman/ so runs can be compared. Katman Pro is $99 a month for up to 3 sites, with fix prompts and page briefs uncapped. Or have us do it: Growth writes 30 pages a month for you. See what Katman does.
What it can’t do: it doesn’t edit the site by itself (your coding tool does), doesn’t publish to WordPress, doesn’t track daily rankings, and costs money where the skills above don’t. If a skill already covers your checks, keep it.
Written by the Katman team, Rnify Limited. GitHub repositories and Claude Code docs read 15 October 2026.
Who this is not for
- Anyone who won’t read the code of what they install: both skills and local MCP servers run with your permissions.
- Teams who need hosted dashboards for many users: neither a skill nor a local server is a shared dashboard.
- Sites that already pass every crawler and rendering check: your next work is pages and mentions, not another audit.
Frequently asked questions
What is a Claude Code skill?
A folder with a SKILL.md file: instructions in Markdown with a short header, plus optional scripts. Claude Code loads it when the task matches its description, or when you type its name as a slash command.
Is geo-seo-claude any good?
It covers a lot for no charge: crawler rules, llms.txt, schema, citability, brand mentions and PDF reports. Read its scripts before running them, and note that its README describes robots.txt checks rather than live requests as each bot.
Can a skill measure whether ChatGPT recommends me?
Only if its scripts call the assistants’ APIs with your keys and keep the answers. The two skill sets we read on 15 Oct 2026 don’t describe that; one tracks brand mentions on platforms such as Reddit and YouTube instead.
Do skills work in Cursor?
SKILL.md is a Claude Code format. Some installers copy skills into other agents, but an MCP server is the standard way to give Cursor, VS Code or Codex the same tools.
Is an MCP server safer than a skill?
Not by itself. Both run on your machine with your permissions; read the code or the permissions page of whatever you install, and keep keys out of committed files.
Can I use a skill and an MCP server together?
Yes. A skill can tell Claude Code when to call the server’s tools, and the server returns the same structured result every time.